NVIDIA CEO, Jensen Huang: "We already have self-improving agents. Set up the right loop, and every run makes it smarter without you." This is the shift that's going to define the rest of 2026. 53 minutes of pure insight from one of the richest men on earth. Watch it,... then read the full guide on building self-improving loops below.show more

@JensenHuang Agentic AI is evolving faster than most people realize.

@JensenHuang Just watched it. The idea of agents that improve without human intervention is wild — we're moving from tools to systems

@JensenHuang Self-improving loops already exist in narrow domains. The harder problem remains designing evaluation systems that keep the improvement aligned with the intended objective

@JensenHuang Will watch the whole clip later today

@JensenHuang jensen huang dropped a ton of great quotes in this video that are worth your attention too his best interview

@JensenHuang The "set up the right loop" part is doing a lot of heavy lifting here. Most self-improving agent systems i've seen hit a ceiling fast because the feedback signal drifts from the actual objective. Jensen's framing skips over that entirely.

@JensenHuang did you ever made a complex agent btw mate?

@JensenHuang The long-term advantage won’t come from smarter models. It will come from systems that continuously improve themselves.

@JensenHuang virtuals protocol. scaled in last friday.watching the flows on arkham.

@JensenHuang The biggest opportunity isn't self-improving agents. It's self-improving workspaces, where every verified outcome strengthens shared Workspace Intelligence.

@JensenHuang A gate caught a bad interval in our copy, and the loop wrote itself a rule that same cycle. Four cycles later, same night, a fresh draft made the identical error. The rule it wrote didn't stop it. An outside gate did, both times. "Without you" is the part I'd test hardest.

Jensen is right. Self-improving agents are just Layer 1. The real breakthrough is self-evolving systems that learn from every execution while remaining governed, auditable, and trustworthy. That's the future we're building at @FinipeAi—where learning, orchestration, evidence, and governance continuously improve together. 🚀 @FinipeAi . 🚀

@JensenHuang can't wait to be the dumbest one in my own company 😎

@JensenHuang the shift from static prompts to self improving loops is the same magnitude as going from a calculator to a spreadsheet one gives you the answer once the other keeps solving forever most people are still calculating one query at a time and wondering why they cant scale

@JensenHuang self-improvement sounds easy until feedback gets noisy

@JensenHuang Jensen says self-improving agents need the right loop. Inference is now >50% of compute cost. The loop isn't for the model, it's for your billing cycle

@JensenHuang The goal isn't to make AI smarter every run. It's to build a workflow that learns from feedback instead of repeating the same mistakes

@JensenHuang Data will be key. To improve wrong data with wrong outcomes is not an improvement. Improving on real data and validated benchmarks will be key